File size: 2,703 Bytes
2189fb9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a256aa0
2189fb9
 
 
 
 
 
 
a256aa0
2189fb9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
"""
Adapted from https://huggingface.co/spaces/stabilityai/stable-diffusion
"""

from tensorflow import keras

keras.mixed_precision.set_global_policy("mixed_float16")

import time

import gradio as gr
import keras_cv

from constants import css, examples, img_height, img_width, num_images_to_gen
from share_btn import community_icon_html, loading_icon_html, share_js

# Load model.
weights_path = keras.utils.get_file(
    origin="https://huggingface.co/sayakpaul/kerascv_sd_pokemon_finetuned/resolve/main/ckpt_epochs_72_res_512_mp_True.h5",
    file_hash="10b20bd27912d1da904dafe8c576351c2f373546f446591aeff00d816d701a6e"
)
pokemon_model = keras_cv.models.StableDiffusion(
    img_width=img_width, img_height=img_height
)
pokemon_model.diffusion_model.load_weights(weights_path)

pokemon_model.diffusion_model.compile(jit_compile=True)
pokemon_model.decoder.compile(jit_compile=True)
pokemon_model.text_encoder.compile(jit_compile=True)

# Warm-up the model.
#_ = pokemon_model.text_to_image("Teddy bear", batch_size=num_images_to_gen)


def generate_image_fn(prompt: str, unconditional_guidance_scale: int) -> list:
    start_time = time.time()
    # `images is an `np.ndarray`. So we convert it to a list of ndarrays.
    # Each ndarray represents a generated image.
    # Reference: https://gradio.app/docs/#gallery
    pokemon_model.to("cpu")
    images = pokemon_model.text_to_image(
        prompt,
        batch_size=num_images_to_gen,
        unconditional_guidance_scale=unconditional_guidance_scale,
    )
    end_time = time.time()
    print(f"Time taken: {end_time - start_time} seconds.")
    return [image for image in images]


description = "This Space demonstrates a fine-tuned Stable Diffusion model. You can use it for generating custom pokemons. To get started, either enter a prompt and pick one from the examples below. For details on the fine-tuning procedure, refer to [this repository](https://github.com/sayakpaul/stable-diffusion-keras-ft/)."
article = "This Space leverages a T4 GPU to run the predictions. We use mixed-precision to speed up the inference latency. We further use XLA to carve out maximum performance from TensorFlow."
gr.Interface(
    generate_image_fn,
    inputs=[
        gr.Textbox(
            label="Enter your prompt",
            max_lines=1,
            placeholder="cute Sundar Pichai creature",
        ),
        gr.Slider(value=40, minimum=8, maximum=50, step=1),
    ],
    outputs=gr.Gallery().style(grid=[2], height="auto"),
    title="Generate custom pokemons",
    description=description,
    article=article,
    examples=[["cute Sundar Pichai creature", 40], ["Hello kitty", 40]],
    allow_flagging=False,
).launch(enable_queue=True)